ããã«ã¡ã¯ãHabrïŒ ãã£ã³ãã«#article_essenseã®Open Data Scienceã³ãã¥ããã£ã®ã¡ã³ããŒããã®ç§åŠèšäºã®ã¬ãã¥ãŒãåŒãç¶ãå ¬éããŠããŸãã 誰ãããæ©ãããããåãåãããå Žå- ã³ãã¥ããã£ã«åå ããŠãã ããïŒ
ä»æ¥ã®èšäºïŒ
- å€æ§ãªç£ç£ã«ããã»ãã³ãã£ãã¯ã»ã°ã¡ã³ããŒã·ã§ã³ã®åŠç¿ ïŒãããã倧åŠãäžæµ·å€§åŠã2018幎ïŒ
- TVAEïŒã¡ããªãã¯åŠç¿ã䜿çšããããªãã¬ããããŒã¹ã®å€åãªãŒããšã³ã³ãŒã㌠ïŒStanford Universityã2018ïŒ
- ç¥çµèšå·åŠç¿ãšæšè«ïŒèª¿æ»ãšè§£é
- ãã¥ãŒã©ã«ãããã¯ãŒã¯ã§ã®å£æ» çãªå¿åŽã®å æ ïŒDeepMindãImperial College Londonã2016幎ïŒ
- Deep Image Harmonization ïŒã«ãªãã©ã«ãã¢å€§åŠãAdobeã2017ïŒ
- å éåŸé ããŒã¹ãã£ã³ã° ïŒãœã«ãã³ã倧åŠã2018幎ã¬ã³ã倧åŠïŒ
1.å€æ§ãªç£ç£ã«ããã»ãã³ãã£ãã¯ã»ã°ã¡ã³ããŒã·ã§ã³ã®åŠç¿
èè
ïŒLinwei YeãZhi LiuãYang WangïŒãããã倧åŠãäžæµ·å€§åŠã2018幎ïŒ
âãªãªãžãã«èšäº
ã¬ãã¥ãŒäœæè
ïŒEgor PanfilovïŒin slack egor.panfilovïŒ
ãã®èšäºã§ã¯ãèè ã¯ãã»ãã³ãã£ãã¯ã»ã°ã¡ã³ããŒã·ã§ã³ã®ã¿ã¹ã¯ã§CNNãæããããã®ã¢ãŒããã¯ãã£ãææ¡ããŸããããã«ããããã¬ãŒãã³ã°çšã«ããŸããŸãªã¿ã€ãã®ããŒã¯ã¢ããïŒãã¯ã»ã«ãã¹ã¯ãå¢çããã¯ã¹ãç»åå šäœã®ã©ãã«ïŒãæã€ããŒã¿ã䜿çšã§ããŸãã
ãã®ã¢ãŒããã¯ãã£ã¯ãå ¥åãšããŠãã³ãœã«ïŒh w 3ïŒãšçºè¡ãã³ãœã«ïŒh w ïŒC + 1ïŒïŒãåãåãFCNã§ãã å ã®ç»åãšåã空é寞æ³ã§ã 次ã«ãåºåãã£ãŒãã£ãããã3ã€ã®ç®æšã§äœ¿çšããŠæ倱ãèšç®ããŸãïŒïŒ1ïŒL_imageïŒã°ããŒãã«å¹³åããŒãªã³ã°->ãã«ãã«ããŽãªBCEæ倱ïŒãïŒ2ïŒL_bboxãïŒ3ïŒL_pixelïŒå¹³åãã¯ã»ã«ããšã®BCEïŒã
L_bboxã¯ããã®èšäºã§æãèå³æ·±ãéšåã®1ã€ã§ãã ããã§ã®åé¡ã¯ããªããžã§ã¯ãã®bboxãæ倱ã¹ã«ã©ãŒã«å€æããæ¹æ³ã§ãã èè ã®ã¢ãããŒãã¯ãbboxã«åºã¥ããŠãªããžã§ã¯ãã®è¿äŒŒãã¹ã¯ãçæããããšã§ãïŒGrabcutãMCGãåçŽãªbboxå¡ãã€ã¶ããããã³UCMïŒultrametric contour mapïŒãè©Šè¡ãããŸããïŒã UCMã¯äœãããããŸãæ©èœããŸãããã€ãŸããè€æ°ã®bboxããªãŒããŒã©ãããããšãã«ããã¯ã»ã«ãåã¯ã©ã¹ïŒsoft.segmãïŒã§èæ ®ãããããŒãžã§ã³ã§ãã ããã1ã€ã®ã¯ã©ã¹ïŒhard.segmãïŒã«å²ãåœãŠãããŠãããªãã·ã§ã³ã¯ãããèªäœãå°ãæªãããšã瀺ããŠããŸãã UCMãã¹ã¯çæã¯æ¬¡ã®ããã«æ©èœããŸãïŒïŒaïŒUCM茪éã®éžæãïŒbïŒåŒ·åºŠã®æ£èŠåãïŒcïŒãããå€ã®3ã€ã®å€ã®èšå®ïŒ1 / 4ã2 / 4ã3 / 4ïŒãïŒdïŒåå¿å茪éã®å¡ãã€ã¶ãã®éå§äžå®ã®å²åã§ãšãªã¢ãé¢ãããŸã§ã ãããã®ãŽãç®±ã®ãã©ã€ããŒããã¹ã¯ã®æ®ãã«äžç¢ºå®æ§ã©ãã«ãå²ãåœãŠãŸã-æ倱ãèšç®ãããšããããããç¡èŠããŸãã ãããã¯ãŒã¯äºæž¬ãšäœæãããåç §ãã¹ã¯ã䜿çšããŠãæ倱ãèæ ®ããŸãã
å®éšã«ã€ããŠïŒèè ã¯ãPASCAL VOC 2012ã§FCNãšDeepLabããããŸããŸãªã¿ã€ãã®ã¬ã€ã¢ãŠãã䜿çšããŠãããŸããŸãªæ¯çïŒãã¹ã¯ïŒbboxïŒã©ãã«-1ïŒ1ïŒ1ãã1ïŒ5ïŒ10ïŒã§ãã¬ãŒãã³ã°ããŸããã çµæã¯ãbboxãéåžžã«åªããŠããããšã瀺ããŠããŸãããéåžžã«é žæ§ã®ããŒã¯ã¢ããïŒãã¯ã»ã«ãã¹ã¯ãéåžžã«å°ãªãå ŽåïŒã䜿çšãããšãç»åã¬ãã«ã®ã©ãã«ã圹ç«ã¡ãŸãã
2. TVAEïŒã¡ããªãã¯åŠç¿ã䜿çšããããªãã¬ããããŒã¹ã®å€åãªãŒããšã³ã³ãŒããŒ
èè
ïŒHaque IshfaqãAssaf HoogiãDaniel RubinïŒã¹ã¿ã³ãã©ãŒã倧åŠã2018幎ïŒ
âãªãªãžãã«èšäº
ã¬ãã¥ãŒäœæè
ïŒEgor PanfilovïŒin slack egor.panfilovïŒ
ãã®èšäºã§ã¯ãã¹ã¿ã³ãã©ãŒãããã®æ°é®®ãªè¡ããã£ãŒãã¡ããªãã¯ã©ãŒãã³ã°ã®åéã«è²¢ç®ããããšããŠããŸãã ãªããäœãå®éã«åé¡ã解決ããããšããŠããã®ããšãã質åã«çããã«ãèè ã¯åçã«VAEã䜿çšããããããã¯ããå¹³åãã¯ãã«ãåŒãåºãïŒåæ£ã«æ³šæãæããïŒãããã«ããã3åã®æ倱ã«å·»ã蟌ãããšãææ¡ããŸãã ãããã£ãŠããããã¯ãŒã¯ã¯ãL_tv_ae = L_reconstruction + L_KL + L_tripletã®æ··åè»è·¡ã§ãã¬ãŒãã³ã°ãããŸãã
圌ãã¯ãMNISTã§ã20ã®ç解ã§ããªããããã¯ãŒã¯ã«åã蟌ã¿ã®æ¬¡å ããç解ã§ããªããã©ã¡ãŒã¿ãŒã§ãç解ã§ããªãæéã§ãã¬ãŒãã³ã°ããŸãã 圌ãã¯ãŸããæªéãã©ã®ããã«ç¥ã£ãŠããããè©äŸ¡ããŸã-ããªãã¬ããæ倱ããŒãã§ããããªãã¬ããã®å²åããã¬ãŒãã³ã°åŸã«èšç®ããããšã«ãã£ãŠã ãã®ãå質ææšãã®ææšã¯æ¹åãããŠãããæ倱é¢æ°ã«ãã®éããŒãžã§ã³ãè¿œå ããããããæå°åããããšãããããäžè¬çã«äºæ³ãããŠããŸãã æåŸã®æ³šæç¹ã¯ãäžæãªæ··ä¹±ã䌎ã2ã€ã®ã¢ãããŒãïŒVAEãšVAE +ããªãã¬ããïŒã®t-SNEåã蟌ã¿ã§ãã ã¯ã©ã¹ã¯ã©ã¹ã¿ãŒïŒèŠããšããïŒã¯ããã³ã³ãã¯ãã«èŠããŸããã1ã€ã®ã¯ã©ã¹ãè€æ°ã®ã¯ã©ã¹ã¿ãŒã§è¡šãããå Žåãããªãå°ãªããªããŸãã
3.ãã¥ãŒã©ã«ã·ã³ããªãã¯åŠç¿ãšæšè«ïŒèª¿æ»ãšè§£é
èšäºã®èè
ïŒBesoldãGarcezãBaderãBowmanãDomingosãHitzlerãKuehnbergerãLambãLowdãLimaãde PenningãPincasãPoonãZaverucha
âãªãªãžãã«èšäº
ã¬ãã¥ãŒäœæè
ïŒEvgeny BlokhinïŒ ebt ïŒ
ãã®ã¬ãã¥ãŒã§ã¯ããã¥ãŒã©ã«ã·ã³ãã«ã³ã³ãã¥ãŒãã£ã³ã°ã«ã€ããŠãAIãžã®2ã€ã®ã¢ãããŒããçµã¿åãããããšããè©Šã¿ãšããŠèª¬æããŠããŸãã 58ããŒãžãããéåžžã«åé·ã§ãã2ã€ã®éšåã§æãçãèŠçŽã瀺ããŸãã ããŒã1.ïŒç·šéæžã¿ïŒã
äžæ¹ã§ãããžãã¯ïŒèšå·ïŒã¯ãã³ã³ãã¥ãŒã¿ãŒãµã€ãšã³ã¹ã®èšç®ãã§ãããäœå幎ãã®éããããŒã«ã«ãŒã«ãã®ããžãã¯ã¯ç©æ¥µçã«éçºãããŠããŸããïŒGOFAIãåç §ïŒã ä»æ¥ãäžæçãéå調ãèšè¿°çããã³ãã®ä»ã®ã¿ã€ãã®ããžãã¯ããããŸãã ç¥è管çã¯æ±ºå®è«çãªããã»ã¹ã§ãã äžæ¹ãæ©æ¢°åŠç¿ïŒã³ãã¯ã·ã§ããºã ïŒã¯çµ±èšçãªæ§è³ªãæã£ãŠããŸãã éåžžã«äžè¬çãªæå³ã§ã¯ããããã¯2ã€ã®åçã®ã¢ãããŒãã§ãããããã¯èšç®å¯èœã§ãããæéãªãŒãããã³ã§è¡šãããšãã§ããŸãã ãã¥ãŒã©ã«ãããã¯ãŒã¯ã¯æ°ããã¢ãããŒãã§ã¯ãããŸããã äžè¬ã«ãåºç¯ãªè¡šçŸåã®æšè«ã¯ããã¥ãŒã©ã«ã·ã³ããªãã¯ãããã¯ãŒã¯ã«ãã£ãŠå®è¡ã§ããŸããããã§éèŠãªã®ã¯ã¢ãžã¥ãŒã«æ§ã§ãã
ããšãã°ãå°æ¬ããããã³ãã³ãšåœŒã®æ·±ããã¬ãŒãã³ã°ã«ã€ããŠèšãã°ãã·ã³ããªãã¯ç¥èã«é¢äžãããµãããããçžäºã«ååž°çã«åœ¢æãããå Žåããã¥ãŒã©ã«ã·ã³ãã«ãããã¯ãŒã¯ãç·ç¶åïŒç·ç¶åããã³ãã«ïŒãšããŠæ³åã§ããŸãã å€æ°XãYãããã³ZãšãPïŒXãYïŒããã³QïŒZïŒã®æŠå¿µãæ åœãã2ã€ã®ãµãããããæ³åããŠãã ããã 次ã«ããã¡ã€ããªã³ã°ã®ã¡ã¿ãããã¯ãŒã¯ã¯ãPïŒXãYïŒ^ QïŒZïŒ-> RïŒXãYãZïŒãšãªãããã«ãPãšQãæ°ããæŠå¿µRïŒXãYãZïŒã«ãããã³ã°ããŸãã å®éã®å®è£ ã¯éåžžã«å°é£ã§ããã èªç¥ã¢ãã«ã¯ã芳枬å¯èœãªããŒã¿ã®è€éãªé¢ä¿ãåŠçããå¿ èŠããããŸãïŒæŠè¡çæ瞊ãå®å šé転ãªã©ïŒã æ®å¿µãªãããå°é家ã¯ãããé決å®çãã€äž»èŠ³çã«ïŒããšãã°ãã¹ãã¬ã¹ãç²åŽã«åºã¥ããŠïŒè¡ãããšãå€ãããã®çµéšãã¢ãã«ã«äŒããæ¹æ³ã¯æ確ã§ã¯ãããŸããã
ãã1ã€ã®äŸã¯ãSutskeverãHinton and TaylorïŒ2009ïŒã®Boltzmannã®ååž°çæéå¶éãã·ã³ïŒRBMïŒã«åºã¥ãNSCAïŒNeural Symbolic Cognitive AgentïŒã¢ãŒããã¯ãã£ã§ãã 2å±€RBMãããã¯ãŒã¯ã¯ãéã¿ãå€æŽããããšã§æçžè«çã®åœ¢åŒã§ããã€ãã®ç¥èãåãåããå ¥åãã¯ãã«ã«å¿ããéã¿ã«ããåºåéã®èšç®ã¯ç¢ºçè«çè°è«ã§ãã ãããã£ãŠãæçžè«çèŠåã¯RBMã®éã¿ã«ãã£ãŠå®çŸ©ãããŸãã NSCAã¢ãŒããã¯ãã£ã¯ãé転ã·ãã¥ã¬ãŒã¿ãDARPA Mind's Eyeãããžã§ã¯ããããã³Visual Intelligenceã·ã¹ãã ã§äœ¿çšãããŸããã ãŸãã2014幎ã«ã¯ãèªåè»ã®CO2åæžã·ã¹ãã ã«é¢äžããŸããã NSCAã«ã¯ãŸã å€ãã®æªè§£æ±ºã®åé¡ãããããšã瀺åãããŠããŸãããçŸåšã¯HintonãšSalakhutdinovã®ãã£ãŒããã«ããã³ãã·ã³ãšçµ±åãããŠããŸãã
ç¥çµèšå·ã·ã¹ãã ã¯ãèªç¥ç§åŠãšç¥çµç§åŠã«ç±æ¥ããŸãã 人éã®æèã®æåã®çè«ã¯ã1983幎ã«ãžã§ã³ãœã³ãšã¬ã¢ãŒãã«ãã£ãŠçå®ãããŸããããããŸã§ãäžå¿çãªåé¡ã¯ãç¥çµçµåã®ã¡ã«ããºã ïŒçµååé¡ïŒã§ãã åé¢ãããç»åãã©ã®çšåºŠæ£ç¢ºã«é©åãããã ããšãã°ããã¥ãŒã©ã«ãªã³ã¯ã®æ¥ç¶ã³ãŒãã®çè«ã¯ãæ¢åã®äººå·¥ãã¥ãŒã©ã«ãããã¯ãŒã¯ïŒANNïŒã«æãè¿ãã§ãã ãã®çè«ã«ãããšãè³å ã®ãã¥ãŒãã³ã«ã¯åäžã®çµåãªãœãŒã¹ïŒçµã¿åããççºïŒã¯ãããŸããããã¹ã±ãŒã©ãã«ãªåæ£è¡šçŸããããããã«ãå¶çºçã«çºçããå¹æçã«èå¥ããããã®ããããŸãã ããããç¥çµç§åŠã¯äžè¬ã«ãçŸä»£ã®ANNãæåŠããŠæèã説æããŸããããã®é·æã¯èªããŠããŸãã ã ããã1988幎以æ¥ãFodorãšPylyshynã¯ãæèã¯æ¬è³ªçã«è±¡åŸŽçã§ãããšäž»åŒµããŠããããœãããŠã§ã¢ãšãââãŒããŠã§ã¢ãšã®é¢çœãé¡äŒŒæ§ãåŒçšããŠããŸãã æ©èœçãªè€éãã¯ãœãããŠã§ã¢ã«ãããæ¬è³ªçã«èšå·ã§ãããã¢ã«ãŽãªãºã ã¯ããŒããŠã§ã¢ã§ã®è¡šçŸã§ã¯ãªããœãããŠã§ã¢ã«ãã£ãŠæ±ºå®ãããŸãã ããã«ãçŸä»£ã®å®éšã§ã¯ã人ã¯èªç¥åé¡ã解決ããããã«ã«ãŒã«ã䜿çšããåŸåãããããããã®ã«ãŒã«ã¯åå¥ã§ãããIQãšçžé¢ããŠããããšã瀺ãããŠããŸãã ããã«ãåé èã®èšç»ãšããŒã«ã©ã€ãºãæ åœããäžå€®å®è¡æ©èœããããŸãã ããã«ãèšèªåœ¢æ ãèŠåã«åºã¥ãã®ãããããšãé£æ³ã«åºã¥ãã®ãã«ã€ããŠãå¿çèšèªåŠè ã®éã§ãŸã è°è«ããããŸãã æåŸã«ãã³ãã¯ã·ã§ããºã ã¯ãå€ãã®èªç¥èœåã®ä»£è¡šçãªæ§æçæ§è³ªãåæ ããŠããŸããïŒãIvan loves MashaïŒãã®ã€ãŽã¡ã³=ãMasha loves Ivanãã®ã€ãŽã¡ã³ïŒã çŸä»£ã®ANNã®æŠå¿µã§ã¯ãäžèšã®åŽé¢ã¯äžæº¶æ§ã§ãã
çŸä»£ã®ANNã®è«ççæšè«ã¯ã次ã®ããã«æ瀺ãããŸãïŒPinkas and Limaã1991-2013ãåç §ïŒã ç§ãã¡ã®ä»äºã¯ãäžæ¬¡è«çã§ç¥èããŒã¹ãæãã察称çãªéã¿è¡åïŒããšãã°ãRBMïŒãæã€ANNãåãããšã ãšèšããŸãã ç¥èããŒã¹ã¯ããã¥ãŒãã³ã®éã¿ãŸãã¯æŽ»æ§åã«ãã£ãŠè¡šãããŸãã 次ã«ãANNã¯ãšãã«ã®ãŒé¢æ°ã®åŸé éäžãå®è¡ããŸããããã®ã°ããŒãã«ãªæå°å€ã¯æšè«ã®é£éã®çµè«ã§ãã
第2éšã§ã¯ãANNã®1次ã®æšè«ãšè«çæŒç®ã®äŸããã«ã³ãè«çãããã¯ãŒã¯ãLSTMã䜿çšããä»ã®ã¢ãããŒãã®äŸãå°æ¥ã®å°é£ãšäºæž¬ã
4.ãã¥ãŒã©ã«ãããã¯ãŒã¯ã§ã®å£æ» çãªå¿åŽã®å æ
èè
ïŒãžã§ãŒã ãºã«ãŒã¯ãããªãã¯ä» ç ïŒDeepMindãã€ã³ããªã¢ã«ã«ã¬ããžãã³ãã³ã2016幎ïŒ
âãªãªãžãã«èšäº
ã¬ãã¥ãŒäœæè
ïŒYana SeredaïŒin slack @ yane123ïŒ
ãã¹ãŠã®ãããã¯ãŒã¯ãã©ã¡ãŒã¿ãåŠç¿ã¿ã¹ã¯Aã®çç£æ§ã«ãšã£ãŠçããéèŠã§ãããšã¯éããªããšããä»®å®ãç«ãŠãŸãããåŠç¿ã¢ã«ãŽãªãºã ãå€æŽããŠãåŠç¿æžã¿ã¿ã¹ã¯ã®ãéèŠãªãéã¿ãç¹å®ããéå»ã®éèŠåºŠã«å¿ããŠåŒ·åãªå€æŽããä¿è·ããããšã決å®ããŸããåŠç¿ããã¿ã¹ã¯ã
ããã¯ãæ倱é¢æ°ã«ããã«ãã£ãè¿œå ããããšã§éæãããŸããã ãã©ã¡ãŒã¿ãŒïŒéã¿ãŸãã¯ãã€ã¢ã¹ïŒã®å€æŽã«å¯Ÿããããã«ãã£ãŒã¯ãå€æŽã®2ä¹ã«æ¯äŸããŠéžæãããŸãã æ¯äŸä¿æ°ïŒã€ãŸããåé¡Aã®ãã©ã¡ãŒã¿ãŒã®éèŠæ§ã®ææšïŒã¯ããã®ãã©ã¡ãŒã¿ãŒãšåé¡Aã®ããŒã¿ã»ããã«å¯ŸããŠèšç®ããããã£ãã·ã£ãŒæ å ±ã§ãã1次å°é¢æ°ã¯ããã£ãã·ã£ãŒè¡åãèšç®ããã®ã«ååã§ãã
ã¿ã¹ã¯AãšBã®åŸãåé¡Cã®ãããã¯ãŒã¯ãåŠç¿ãå§ããå Žåãæ倱é¢æ°ã¯ãããã3é ã«ãªããŸãïŒæå·Aãæå·Bãããã³ãœãªã¥ãŒã·ã§ã³Cã®å質ïŒã
圌ãã¯ãç°ãªãåé¡ã解決ããããã«ããããã¯ãŒã¯ã«ããåãéã¿ã®åå©çšã®çšåºŠã枬å®ããæ¹æ³ãããã£ãã·ã£ãŒã®ãªãŒããŒã©ããããææ¡ããŸããã
å®éšïŒåæãã ãã¹ããã¢ã¿ãªã«ã€ããŠ
5.æ·±ãç»åã®èª¿å
èšäºã®èè
ïŒYi-Hsuan TsaiãXiaohui ShenãZhe LinãKalyan SunkavalliãXin LuãMing-Hsuan YangïŒã«ãªãã©ã«ãã¢å€§åŠãã¢ããã2017幎ïŒ
âãªãªãžãã«èšäº
ã¬ãã¥ãŒäœæè
ïŒArseniy KravchenkoïŒin slack @arsenyinfoïŒ
ããç»åãããªããžã§ã¯ããåãåã£ãŠå¥ã®ç»åã«è²Œãä»ãããšããããã§ã¯ãªããšæšæž¬ããã®ã¯ç°¡åã§ããå ã圱ãªã©ã®éããç®ãåŒããŸãã 圌ãã®æŽ»åãé ãããã«ãPhotoshopãã¹ã¿ãŒã¯ããã€ãã®æ²ç·ãã²ãããŸãããDLã®å°å¹Žã«ã¯ä»£æ¿ãœãªã¥ãŒã·ã§ã³ããããŸãã
貧匱ã«æ®åœ±ãããåçãã»ãŒå®éã®åçã«å€ãããšã³ãããŒãšã³ãã®ãããã¯ãŒã¯ããã¬ãŒãã³ã°ããŸãããã ããŒã¿ã®åé¡ïŒããã§ã¯ãæåž«ããåŠç¿çšã®å€å žçãªXãYãã¢ã¯ãããŸããã ãããããã®ãããªããŒã¿ãçæããããšã¯ã§ããŸãïŒCOCOãªã©ã®ããŒã¿ã»ããããåçãæ®ãããã¹ã¯ã§ãªããžã§ã¯ããéžæããããããå°ç¡ãã«ããã-è²ç¹æ§ã転éããŸãïŒããšãã°ããã¹ãã°ã©ã ãããã³ã°ã䜿çšããŸãããèè ã¯ããé«åºŠãªææ³ã䜿çšããŸããïŒã ãªããªã COCOã®å€æ§æ§ã¯ååã§ã¯ãããŸãããåæ§ã«æªãã ç»åãã¡ãã€ãããåãé€ãããã£ã«ã¿ãŒã§é€å€ããŸãããïŒhelloãïŒproj_kaggle_cameraïŒïŒçŸåŠäºæž¬ã¢ãã«ã«ããããããŸãã
ããã§ããããã¯ãŒã¯ãšã³ã³ãŒããŒãã³ãŒããŒã¢ãŒããã¯ãã£ãåŠç¿ã§ããŸããunetïŒéåžžã®L2æ倱ã䜿çšããŠãæªãã ç»å+æªã¿ãã¹ã¯ãå ã®ç»åã«å€æããŸãã å質ãããã«åäžãããããã«ããã«ãã¿ã¹ã¯åŠç¿ã«åãçµã¿ãŸããã»ãã³ãã£ãã¯ã»ã°ã¡ã³ããŒã·ã§ã³ãæ åœãã2çªç®ã®ãã³ãŒããã©ã³ããè¿œå ããæ倱ã§ã¯ããããã¯ãã¹ãšã³ããããŒãè¿œå ããŸãã çµæã®ã¢ãŒããã¯ãã£ã¯æ¬¡ã®ããã«ãªããŸã
åçã¯æ¬åœã«ã¯ãŒã«ã«èŠããŸãã
èè ã¯ã¢ãã«ãšééãCaffeã«æçš¿ããŸããããèŽåœçãªæ¬ é¥ããããŸã-ç§ã¯å人çã«ã¯ãŸã ã¢ãã«ãæã¡äžããããšãã§ããŸããã§ããïŒç§ã¯caffeã®äœããã®ä¿®æ£ããŒãžã§ã³ãå¿ èŠãªããã§ãããã©ããæ確ã§ã¯ãããŸããïŒã
6.å éåŸé ããŒã¹ãã£ã³ã°
èšäºã®èè
ïŒGérardBiauãBenoîtCadreãLaurentRouvìÚreïŒãœã«ãã³ã倧åŠã2018幎ã¬ã³ã倧åŠïŒ
âãªãªãžãã«èšäº
ã¬ãã¥ãŒäœæè
ïŒArtem SobolevïŒasobolev slack ïŒ
æé©åçè«ã¯ãåŸé éäžãããªãé ãæé ã§ããããšãç¥ã£ãŠããŸãã ãããã ãã¥ãŒãã³ã®æ¹æ³ãããã»è¡åã䜿çšããŠãã©ã¡ãŒã¿ãŒã®å€åã®æ¹åãä¿®æ£ããŸãããèšç®ãããè€éã§ãè¿œå ã®èšç®ãå¿ èŠã§ãã äžæ¬¡ã®æ¹æ³ãããªãã¡ åŸé ãšé¢æ°å€ã®ã¿ã䜿çšããŠãæé©ãªïŒç¹å®ã®ã¯ã©ã¹ã®åé¡ã«å¯ŸããïŒåæçã¯ãNesterovã®é«éåŸé æ³ã§ãã
ãã®èšäºã§ã¯ããã®æé©åææ³ãåŸé ããŒã¹ãã£ã³ã°ã«æ¿å ¥ããããšãææ¡ããŸãã åçŽã§ãããšèããããŸãïŒãã®æ¹æ³ã¯ããéãåæããŸããã€ãŸãããã®ãããªé«éåŸé ããŒã¹ãã¯ãããå°ãªãã¢ã³ãµã³ãã«ã¢ãã«ã§ããè¯ãå質ãæã€ã¯ãã§ãã
èè ã¯ãéåžžã®åŸé ããŒã¹ãã£ã³ã°ãšæ¯èŒããŠãããã€ãã®å®éšãè¡ã£ãã 䜿çšãããããŒã¿ã¯å¥åŠã§ãããããã€ãã®åæããŒã¿ã»ãããš5ã€ã®UCIãªããžããªã å®éãèè ã¯é«éåŸé ããŒã¹ãã£ã³ã°ãéåžžãããã¯ããã«é«éã«ãã¬ãŒãã³ã°ãããããšã瀺ããŠããŸãïŒå³3ïŒããåæããŒã¿ã§ããã瀺ããŠããŸãã äžè¬ã«ã圌ãã®å®éšã®çµæã«ãããšãèè ã¯ããã®æ¹æ³ã¯ããã»ã©ããŸãæ©èœãããåŠç¿çã«å¯Ÿããæ床ãäœããã¢ã³ãµã³ãã«ã§äœ¿çšããã¢ãã«ãå°ãªããšäž»åŒµããŠããŸãã